Identifying and Predicting Cognitive Decline Using Multi-Modal Sensor Data and Machine Learning Approach.

Alzheimer's Disease (AD) remains a critical global health challenge, with its prevalence expected to rise dramatically by 2050, leading to substantial financial and emotional burdens. Mild Cognitive Impairment (MCI), the prodromal stage of AD, presents a crucial opportunity for early intervention, yet its diagnosis remains difficult due to the overlap with normal aging. Traditional diagnostic methods, such as neuroimaging and cerebrospinal fluid analysis, are costly and invasive, highlighting the need for alternative, scalable, and non-invasive biomarkers. This study explores the potential of naturalistic driving behavior as a digital biomarker for detecting cognitive decline in individuals at risk for AD and MCI. A total of 118 participants (8 with AD, 65 with MCI, and 45 cognitively healthy individuals) were included in this study. At baseline year, we measured their demographics, cognitive status administrated by dementia experts, 3 consecutive months of naturalistic driving performance and driving life-space from participants' own vehicle and sleep data via wrist-worn actigraphy, integrated into multi-modal data to feed to XGBoost-based framework. After 1-year follow, their cognitive status was assessed. We implemented a two-phase validation framework: first, classification model using Leave-One-Subject-Out Cross-Validation (LOSO-CV) to classify baseline cognitive status, and then, conducting a prediction model with to assess the model's ability to predict 1-year follow-up cognitive status. Our results demonstrate that the multi-modal classifier achieved strong classification performance (accuracy = 68.64%; precision = 73.97%; F1-score=74.48%), with the highest recall (76.39%) from a model incorporating demographics and driving features, and prediction performance (accuracy = 70.48%; precision = 71.88%; F1-score = 74.80%, recall = 77.97%). Key predictive features included sex, mean awakening duration, age, average acceleration, and sleep efficiency, underscoring the relevance of driving behavior and sleep characteristics in cognitive assessment. By leveraging everyday activities such as driving, this framework provides a novel, non-invasive approach for identifying individuals at risk for cognitive decline. Furthermore, its ability to predict future disease progression establishes a forward-looking paradigm for early detection and monitoring. Beyond cognitive impairment, this methodology offers a scalable and generalizable framework for disease prediction, with potential applications in detecting and monitoring other neurodegenerative and chronic conditions.